Google says Willow chip cut quantum 'magic state' errors about 40-fold
Google researchers say they used the company's Willow quantum chip to cut errors in special "magic states" about 40-fold, New Scientist and Scientific American reported. These states are needed for the most powerful quantum math. Outside experts praised the result but noted that most attempts were thrown out.
Researchers at Google Quantum AI say they have used the company's Willow quantum computer to make "magic states" far more reliably, a possible step toward cheaper error-corrected quantum computers, New Scientist and Scientific American reported. The study was published in the journal Science.
Quantum computers are very error-prone, New Scientist reported. To run the operations that give them their edge, called non-Clifford gates, they need qubits prepared in a precise "magic state." Making those states has been a real bottleneck, lead author Emma Rosenfeld told the magazine. "It eats up a lot of qubits and a lot of time," she said.
The standard method, called distillation, can use up most of a machine, Rosenfeld told Scientific American. "You can have up to 95 percent of your quantum computer being spent just generating these high-fidelity magic states," she said.
The team instead tested a method Google colleagues proposed in 2024, called magic state cultivation, in which the state is checked and then "grown" across more qubits, both outlets reported. Without cultivation, about four attempts in 1,000 produced a faulty state, and with it errors fell by roughly 40-fold, Scientific American reported. The experiment used 54 of Willow's 105 physical qubits, the magazine said.
Outside experts were impressed. "This is an impressive demonstration of a promising recent idea pioneered by Google," Isaac Kim of the University of California, Davis, told New Scientist. "The Google experiment shows that cultivation works in practice," Kenneth Brown, director of the Duke Quantum Center, told Scientific American.
There are limits. Only 8 percent of Willow's runs produced high-quality magic states, Kim noted to New Scientist. Zhenyu Cai of Imperial College London told the magazine that Willow's errors must come down further before it can tackle truly useful problems. The team is at about one error in 10,000 attempts, while breaking RSA-2048 encryption would need one in 10 billion, Scientific American reported.
"This is the first demonstration that the concepts work. We're not done," Rosenfeld told New Scientist.
What we know: Google's team reports a roughly 40-fold drop in magic state errors on Willow, in a study in Science.
What we don't: Whether the method can scale to the far lower error rates useful programs need.
Where outlets agree New Scientist and Scientific American both report that Google's team used magic state cultivation on Willow to cut magic state errors about 40-fold.
Where they differ The outlets cited here report the same key facts so far.